[__jl__]

In our experience, caching is not very reliable with google. We always get random cache misses that don't happen with other providers. We find OpenAI, Anthropic and Fireworks (which we use a lot) all have higher cache hit rates. So it's not only about the costs of cached token but also what kind of cached hit rate you get.

[svachalek]

In my experience Google is the most flaky in general, which is surprising considering the rock solid history of their search and other products. Just more likely not to respond at all, to give a response out of left field, to handle the same error in 12 different ways randomly (a rainbow of HTTP status codes and error messages), etc etc.

[gwern]

I agree. The https://aistudio.google.com/ is shockingly bad. I'm not sure I've ever used such a flaky Google service before. It's so much worse than Gmail or Google, not to mention ChatGPT or Claude or DeepSeek or Kimi or Midjourney web interfaces. The bizarre janky integration with your Google Drive, or Gemini or NBPs randomly erroring out, often indefinitely. I've had sessions refresh themselves and just... disappearing. Or when you get frustrated with a buggy dead session and hit 'new session' and have to wait minutes for 'saving...' to happen.

[veselin]

Exactly our experience too. Effectively we catch these and on these status codes, we send to OpenAI. Retrying the same query in Gemini has high chance to give kind-of the same status code.